Tutorial 3 — Real-Time Pose Estimation with YOLOv8 on Jetson

Pose estimation detects 17 body keypoints per person — nose, eyes, shoulders, elbows, wrists, hips, knees, and ankles — at 40+ FPS. Use it for sports analysis, ergonomics, fall detection, or gesture control.

What you will learn

• How pose estimation works — keypoints vs bounding boxes
• How to detect 17-point body skeletons in real time
• How to calculate joint angles (e.g. elbow bend, knee flexion)
• How to classify actions from pose (standing, sitting, falling)
• Applications: sports coaching, workplace ergonomics, fall detection

Step 1 — Run the pose demo

cd ~/tutorials/03-pose-estimation
python3 pose.py --source 0 --show

You will see a skeleton drawn over every person in frame. Each of the 17 keypoints is shown as a coloured dot, connected by lines representing limbs.

Step 2 — Access keypoint coordinates

from ultralytics import YOLO
model = YOLO("yolov8n-pose.engine", task="pose")
results = model(frame)

for person in results[0].keypoints.data:
    # 17 keypoints, each with [x, y, confidence]
    nose       = person[0]   # [x, y, conf]
    left_wrist = person[9]
    right_wrist= person[10]
    print(f"Nose position: ({nose[0]:.0f}, {nose[1]:.0f})")

Step 3 — Calculate a joint angle

import numpy as np

def joint_angle(a, b, c):
    """Calculate angle at joint b, given points a, b, c"""
    ba = np.array(a) - np.array(b)
    bc = np.array(c) - np.array(b)
    cosine = np.dot(ba, bc) / (np.linalg.norm(ba) * np.linalg.norm(bc))
    return np.degrees(np.arccos(np.clip(cosine, -1.0, 1.0)))

# Example: elbow angle (shoulder → elbow → wrist)
shoulder = person[5][:2].numpy()
elbow    = person[7][:2].numpy()
wrist    = person[9][:2].numpy()
angle    = joint_angle(shoulder, elbow, wrist)
print(f"Elbow angle: {angle:.1f}°")

Step 4 — Simple fall detection

def is_fallen(keypoints):
    """Simple fall detection — head near the same height as hips"""
    nose = keypoints[0]
    hip_l = keypoints[11]
    hip_r = keypoints[12]
    hip_y = (hip_l[1] + hip_r[1]) / 2
    # If head is below hips level, likely fallen
    return abs(nose[1] - hip_y) < 60

✅ Next: Tutorial 4 — AI Plant Health Monitor | Back to Jetson Kit